Abstract
Social networks can serve as a valuable communication channel for asking for help, offering assistance, and coordinating rescue activities in disaster because it allows users to continuously update critical information in the fast-changing disaster environment. This paper presents a novel sequence to sequence based framework for forecasting people's needs during disasters using social media and weather data. It consists of two Long Short-Term Memory (LSTM) models, one of which encodes input sequences of weather information and the other plays as a conditional decoder that decodes the encoded vector and forecasts the survivors' needs. Case studies using data collected during Hurricane Sandy in 2012, Hurricane Harvey and Hurricane Irma in 2017 demonstrate that the proposed approach outperformed the statistical language model n-gram, LSTM generative model, and convolutional neural network (CNN) based model. This research indicates its great promise for enhancing disaster management such as evacuation planning and commodity delivery.
Original language | English |
---|---|
Pages (from-to) | 229-240 |
Number of pages | 12 |
Journal | IEEE Transactions on Big Data |
Volume | 8 |
Issue number | 1 |
DOIs | |
State | Published - Feb 1 2022 |
Keywords
- Concern flow
- Disaster relief
- Hurricane events
- LSTM
- Needs forecasting
- Sequence to sequence model